View
9
Download
0
Category
Preview:
Citation preview
A Numerical Framework for Solidification and ResidualStress Modelling in Metallurgical Applications
Sebastian, Kohlstädta,b,*, Michael Vynnyckyb, Hrvoje Jasakc,d
aVolkswagen AG, Kassel, GermanybKTH Royal Institute of Technology, Stockholm, Sweden
cWikki Ltd, 459 China Works, Black Prince Road, London, SE1 7SJ, United KingdomdFaculty of Mechanical Engineering and Naval Architecture, University of Zagreb,
Ivana Lucica 5, 10 000 Zagreb, Croatia
Abstract
Quality of foundry products depends on the control of the solidification process;
if the phase change is not managed correctly, residual stresses and material
shrinkage results in deteriorating product quality.
Physical modelling of solidification in the context of 3-D Computational
Continuum Mechanics (CCM) is particularly challenging: it requires a unified
fluid-solid model capable of simulating fluid flow, heat transfer, solidification
phase change as well as solid mechanics aspects of the resulting material.
In this paper, a unified physical model of solidification is presented. The
model is capable of following the progress of the melt from injection/pouring
of the molten phase into a batch or a continuous casting system to the final
solid part. Governing equations resolve heat transfer, flow of the liquid phase,
solidification phase change with the release of latent heat, and accumulation of
residual stresses in the solid. This is achieved by solving the basic conservation
equations (mass, momentum and energy) with a set of unified liquid-solid con-
stitutive relations covering the range of phase change: flow of an incompressible
non-Newtonian liquid, solid-liquid mixture and large deformation solid stress
model in the incremental form. Numerical formulation of the model incorpo-
rates the effects of heat release due to phase changes, material compressibility,
∗Corresponding authorEmail address: skoh@kth.se (Sebastian Kohlstädt)
Preprint submitted to Ocean Engineering January 22, 2019
plastic yielding and thermal stresses.
The solidification model is implemented in OpenFOAM and used to simulate
transient 3-D casting process. It is validated on limiting cases of fluid flow,
mechanical and thermal stresses solids, conventional fluid-solid interaction and
fluid-to-solid phase transition.
The model is capable of predicting shrinkage and solidification effects and
can be used in optimisation of casting systems, mould pressurisation and heat
transfer management in foundry applications.
Keywords: Fluid-Solid Interaction, Unified Approach, Solidification, Residual
Stresses, FVM, OpenFOAM
1. Introduction
While Computational Fluid Dynamics (CFD) and, more widely, Computa-
tional Continuum Mechanics (CCM) is applied across a range of engineering
applications, solidification phase change in foundry applications are managed
mainly based on operator’s experience. The process of solidification under in-5
tensive heat transfer poses a challenge because it bridges the traditional split of
continuum mechanics models for solids and fluids.
Majority of Fluid-Structure Interaction (FSI) solution methods [1, 2] can-
not handle the problem of an evolving fluid-to-solid interface. Partitioned FSI
models [3] use separate equation sets (and sometimes separate discretisation pro-10
cedures) for the fluid and solid regions, thus a-priori precluding the extension
of the model to solidification. There exists in the literature a class of monolithic
FSI models [4, 5, 6, 7] where a single set of governing equations covers both
phases. In essence, this is trivial: conservation laws for mass, momentum and
energy readily serve this purpose. However, the problem lies in the constitutive15
law which needs to cover both the fluid and the solid phase.
The monolithic FSI approach provides the most promising way forward in
simulation of solidification. Further challenges exists; most notable being the
formulation of fluid models capable of dealing with significant non-Newtonian ef-
2
fects, a solid stress model which includes the handling of geometric and material20
non-linearity (large deformation; plasticity) and adequate handling of thermally-
induced stresses. Particular challenge in the formulation is the handling of a
moving fluid-to-solid interface and the handling of the material undergoing so-
lidification where a solid and liquid phase co-exist.
The primary objective of this study is the simulation of cast shrinkage and25
residual stresses. To achieve this, it is necessary to follow the development of
stress history from the point of solidification to final state, accounting for ther-
mal stresses in the solid and temperature-dependent material properties. In
solidification, a mushy region is assumed, with material properties of the solid
shell and liquid melt dependent on temperature, and the progress of solidifica-30
tion is followed by a liquid fraction variable.
This paper describes a unified fluid-to-solid model of solidification and resid-
ual stress accumulation in metallurgical applications. The challenge is to a as-
semble of a unified model covering both the liquid and solid phase with sufficient
fidelity to capture the thermal front and the stress state at the point of solidi-35
fication, which is the main source of residual stresses, volumetric imperfections
and other quality-degrading features of the cast.
REWRITE: PAPER LAYOUT
In what follows, we shall present the choice of appropriate working variable,
conservation laws and constitutive relations covering the full range of solidifi-40
cation physics. The new model shall be validated on canonical cases of solid
mechanics, fluid flow and conventional fixed interface fluid-solid interaction, in
preparation for solidification studies.
The paper is completed with validation cases TEXT MISSING
3
2. Governing Equations45
The derivation of the unified solidification model is based on the governing
equations for mass, momentum and energy, applicable in both fluids and solids:
∂ρ
∂t+∇•(ρu) = 0, (1)
∂(ρu)
∂t+∇•(ρuu) = ρf +∇•σ, (2)
∂(ρe)
∂t+∇•(ρeu) = ρQ+∇•q+ ρf•u− p∇•u+∇•(σ•u), (3)
where ρ is the density field, u is the velocity field, σ is the material stress tensor,
f is the body force, e is the internal energy, Q is the volumetric energy source
and q is the heat flux.
The greatest challenge in the derivation of the combined fluid-solid model
shall be in the formulation of the unified constitutive relation in the stress tensor50
σ to cover both the fluid and solid to sufficient level of accuracy. Further chal-
lenges, such as the choice of solution variable and equation coupling / solution
algorithm shall follow.
3. Minimal Solidification Model
In this section, a minimal unified form of the solidification model shall be55
considered. Modelling issues such as combined closure formulae and choice of
working variable shall be made first; to be followed by the necessary model
extensions for unified fluid-solid interaction and ultimately solidification.
3.1. Fluid Stress Model
A fundamental difference between fluids and solids is related to the way the
material accumulates stress. In fluids, stress in material is a result of velocity
gradient, as stated by the generalised form of the Newton’s law of viscosity:
σf,l = −
(
p+2
3µl∇•u
)
I+ µl
(
∇u+ (∇u)T)
, (4)
4
where σf,l is the material stress tensor for the linear Stoksean fluid, p is its60
spherical part (pressure), µl is the dynamic viscosity of the and u is the flow
velocity field.
For the fluid, the continuity equation reads:
∂ρl
∂t+∇•(ρl u) = 0, (5)
where ρl i the liquid density. In the incompressible limit, (ρl = const.), Equation 5
is written as:
∇•u = 0. (6)
The fluid momentum equation is derived from the momentum conservation,
Equation 2 and the Newton’s viscosity law, Equation 4:
∂(ρlu)
∂t+∇•(ρluu)−∇•
[
µl
(
∇u+ (∇u)T)]
= −∇
(
p+2
3µ∇•u
)
. (7)
Assuming ρl = const., neglecting the body force term and using Equation 6,
Equation 7 reduces to its incompressible form, yielding the simplest form of the
incompressible fluid flow model:
∂u
∂t+∇•(uu)−∇•(νl∇u) = −
1
ρl∇p, (8)
∇•u = 0.
3.2. Solid Stress Model
In solids, material accumulates stress with the gradient of displacement d
rather than velocity u. The simplest, small deformation linear elastic model
states:
σs,l = 2µsεs + λs tr(εs,l) I, (9)
where σf,l is the material stress tensor for the linear elastic solid, µs and λs
are the Lame’s coefficients, defined in terms of the solid modulus of elasticity E
and is the Poisson’s ratio νs and εs,l is the solid strain defined in terms of solid
displacement d:
εs,l =1
2
[
∇d+ (∇d)T]
. (10)
5
The Hooke’s law, relating the stress and strain tensors, closes the system of
equations:
σs,l = 2µsεs,l + λs tr(εs,l) I, (11)
where I is the unit tensor and µs and λs are the Lame’s coefficients, relating to
the Young’s modulus of elasticity E and the Poisson’s ratio ν as:
µs =E
2 (1 + ν)(12)
and
λs =
ν E(1+ν)(1−ν) for plane stress,
ν E(1+ν)(1−2 ν) for plane strain and 3-D.
(13)
The above can be combined to state the solid stress directly as a function of
displacement d, within the linear approximation:
σs,l = µs
(
∇d+ (∇d)T)
+ λsI tr(∇d). (14)
Neglecting the convective effects in the solid and body force, the simplest
form the linear solid momentum balance with displacement d as a working
variable reads:
∂2(ρsd)
∂t2−∇•
[
µs
(
∇d+ (∇d)T)
+ λsI tr(∇d)]
= 0, (15)
where ρs is the solid density. Equation 15 shall be referred to as the Solid Linear
Elastic Displacement model (SLED), as described in [11]. This model is clearly65
not practical for the unified solid-fluid formulation and further modifications to
the formulation shall be performed below.
3.3. Choice of the Working Variable
The natural choice of a working variable in solids is displacement d or dis-
placement increment δd and the stress-strain tensor pair (when accounting for70
material or geometrical non-linearity), while in fluids the working variable are
regularly velocity u and pressure p.
6
We shall first resolve the problem of displacement. Comparing the expres-
sions above, one can claim that the solid accumulates displacements and there-
fore stresses, while the fluid only reacts to velocity gradients. In the framework
of a discrete time-advancing numerical simulation with the time-step of ∆t,
accumulation of displacement from velocity (or “displacement increment”) is
evaluated as:
dnew = d0 +
∫ t
0
u dt
≈ dold + u∆t, (16)
where, in the context of second-order accurate time-marching temporal inte-
gration dold and dnew are the old- and new-time-level displacements. In this
manner, d is accumulated during the simulation without the need to refer to75
the initial configuration d0.
Blending of the two models is performed using the liquid fraction variable
α, combining governing equations for the fluid and the solid phase. The fluid
model is recovered for α = 0, while the other extreme of α = 1 indicates solid
behaviour. For the moment, we shall insist on a binary α distribution, where80
values other than α = 0, α = 1 are inadmissible.
Consequently, the combined fluid and solid stress for a Newtonian fluid and
linear elastic solid σl is defined using the solid fraction α as:
σl = ασs,l + (1− α)σf,l, (17)
To achieve this, it is necessary to use Equation 16 to reformulate the solid
stress Equation 14 in terms of velocity u.
In preparation for further model development, Equation 14 can be rewrit-
ten to directly accumulate the solid stress, just as the solid displacement is
accumulated from the velocity, Equation 16:
σs,l = σs,l,0 +
∫ t
0
[
µs
(
∇u+ (∇u)T)
+ λsI tr(∇u)]
dt,
≈ σolds,l +∆t
[
µs
(
∇u+ (∇u)T)
+ λsI tr(∇u)]
. (18)
7
where σs,l,0 is the initial stress distribution and σolds,l is the stress from the
previous time-step.85
The combined stress model appears feasible, but suffers from several serious
problems. The first is the presence of the pressure field as a solution in the
fluid and its absence in the solid stress model, which would lead to problems
of force balance at the fluid-solid interface. The second (in preparation to the
solidification modelling), is the failure of the the solid stress formulation at the90
incompressibility limit. The third is the lack of the solid continuity equation
which can be matched with Equation 6 in the combined model.
To overcome these difficulties and before reaching the final form, further
manipulation of the stress model for a linear elastic solid is required.
3.4. Papadakis Solid Pressure Term Formulation95
The linear solid stress model suffers from a particular failure mode, where
the Poisson’s ratio ν reaches the value of 0.5: the second Lame coefficient λs
tends to infinity.
λs =E
3(1− 2ν); ν = 0.5 → λs → ∞, (19)
Note that realistic solids are never considered incompressible, meaning that
λs remains bounded.
While ν = 0.5 may be unrealistic for solids, where it describes the state of
incompressibility, a case of incompressible liquid will certainly be encountered.
Therefore, the solid model failure for ν = 0.5 must be circumvented.100
The basis of model reformulation is the introduction a meaningful pressure
variable in the solid, with an option handle the volumetric change via the solid
continuity (density) equation. At the same time, this resolves the problem of
compatibility of normal stresses at the solid-fluid interface mentioned above.
The pressure manipulation is introduced by Papadakis [5] into the∇•[λsI tr(∇d)]105
term in Equation 18, recognising that the pressure p is related to the trace of
the stress tensor and divergence of the displacement field ∇•d.
8
Manipulation of the divergence terms into the desired form, which matches
the desired definition of the solid pressure to match the fluid p reads:
λstr(∇d) = −
(
p+2
3µs∇•d
)
. (20)
Definition of the solid pressure shall be obtained by rewriting Equation 20 to
express p. Introduction of the solid pressure term also yields a solid continuity
equation and the bulk modulus K:
K = ρ∂p
∂ρ=
2
3µs + λs =
E
3(1− 2ν), (21)
defining the pressure (in the solid region!) as:
p = −1
3tr(σs,l) = −K∇•d. (22)
Note that the bulk modulusK still suffers from the problem of tending to infinity
as the Poisson’s ratio tends to 0.5, which represents the incompressibility limit
(for the solid). This, in turn is not physically realistic as all solids respond to110
external (pressure) loading by changing their volume.
One can express this change via the solid continuity equation, directly from
Equation 22:
∇•d = −1
Kp. (23)
A realistic change of the solid density is negligible as the volumetric change
of the solid under pressure loading is minimal compared to its mean value. To
illustrate this, consider the order-of-magnitude of the volumetric change of a
solid block under pressure loading, compared to the “mean” value of its density.115
The solid will react to the outside loading via deformation instead. One can
therefore linearise out the fact that ρs needs to vary with d and indeed p.
Equation 20 and Equation 14 yield the “pressure form” of the stress equation
for the linear elastic solid in terms of d and p:
σs,l = −
(
p+2
3µs ∇•d
)
I+ µs
(
∇d+ (∇d)T)
. (24)
Note that the meaning of the pressure is identical both in the solid and fluid:
it represents the spherical part of the stress term, as originally intended.
9
Equation 24 allows us to rewrite the solid momentum equation in terms of
pressure
∂2(ρsd)
∂t2−∇•
[
µs
(
∇d+ (∇d)T)]
+2
3∇(µs ∇•d) = −∇p. (25)
Equation 24 is more convenient for stress blending between the fluid and120
solid, Equation 17. In addition, this introduces the solid continuity Equation 23,
which neatly matches the fluid model formulation.
The second formulation of the linear elastic solid model consists of the mo-
mentum Equation 25 and continuity Equation 23:
∂2(ρsd)
∂t2−∇•
[
µs
(
∇d+ (∇d)T)]
+2
3∇(µs ∇•d) = −∇p,
∇•d = −1
Kp
and shall be referred to as Solid Linear Elastic Pressure Displacement model
(SLEPD).
Moving beyond the work of Papadakis, the above equation set, Equation 23125
and Equation 24, can be rewritten in terms of solid velocity (or “displacement
increment”).
The solid continuity equation is derived by taking the time-derivative of
Equation 23:
∇•u = −1
K
∂p
∂t. (26)
For the solid stress, following Equation 18, but in the pressure formulation
Equation 24
σs,l = σ∗
s,l,0 +
∫ t
0
(
µs
(
∇u+ (∇u)T)
−2
3µs∇•u I
)
dt− pI,
≈ σ∗,olds,l +∆t
(
µs
(
∇u+ (∇u)T)
−2
3µs∇•u I
)
− pI, (27)
where σ∗
s,l,0 is the initial stress tensor without the pressure term:
σ∗
s,l,0 = µs
(
∇d0 + (∇d0)T)
−2
3µs ∇•d0 I (28)
and σ∗,olds,l is its accumulated value. For a linear elastic material, this can also
be calculated from d:
σ∗,olds,l,0 = µs
(
∇d+ (∇d)T)
−2
3µs ∇•d I. (29)
10
While the momentum equation for the solid formally contains the convection
term ∇•(ρuu), in a solid material the deformation velocity is so low that the
convective effect can be eliminated.130
Introducing the kinematic Lame’s coefficient νs
νs =µs
ρs, (30)
(NOTE: ν is the Poisson’s ratio and νs is the kinematic Lame’s coefficient,
defined above!)
the equations describing the stress state for a linear elastic solid read:
u =∂d
∂t,→ dnew = dold + u∆t (31)
∂u
∂t−∇•
[
νs∆t(
∇u+ (∇u)T)]
+2
3∇(νs∆t∇•u)
−∇•
[
νs(
∇d+ (∇d)T)]
+2
3∇(νs ∇•d)
= −1
ρ∇p,
(32)
∇•u = −1
K
∂p
∂t.
This model shall be referred to as Solid Linear Elastic Pressure Velocity model
(SLEPV).
Note that the convection term in the solid momentum equation, Equation 32,135
formally exists, but is dropped from implementation: the linear elastic model
does not allow for large deformation where the convective effect in the solid
would be significant.
4. Linear Solid-Fluid Interaction
The model formulation up to this point is sufficient for unified modelling of140
fluid-solid interaction involving a Newtonian fluid and a linear elastic solid.
Looking at the equation set for the fluid, Equation 8, Equation 6 and the
solid, Equation 31, Equation 32, Equation 26, it is clear that they can be di-
rectly blended using the solid fraction α.
11
4.1. Viscous Stress Balance at the Solid-Fluid Interface145
There remains a question of the tangential (“viscous”) stress balance at the
fluid-solid interface. Comparing the fluid and solid model, the stress terms
balance at the interface faces where solid fraction α undergoes a step change
(liquid cell adjacent to solid cell), the shear force balance is achieved between
one term at the fluid side depdending on u and two terms at the solid side,150
one depending on u and another depending on d. Therefore, at the solid-fluid
interface, the velocity Laplacian exhibits a jump condition matched by the the
displacement stress terms in the solid.
The solid model formulation has purposefully introduced the solid pressure,
indicating that the continuous pressure distribution at the fluid-solid interface155
indicates normal stress balance without further intervention.
4.2. Linear Solid-Fluid Interaction Model
For completeness, the linear solidification model is derived by blending the
fluid and solid model, using the solid fraction variable α which assumes a step-
change at the fluid-solid interface using Equation 17.160
Solid displacement integration equation follows from Equation 31, account-
ing for the phase fraction α and reads:
dnew = d0 +
∫ t
0
αu dt
≈ dold + αu∆t. (33)
Equivalently, for the solid stress, based on the pressure form, Equation 27:
ασs,l = ασ∗
s,l,0 +
∫ t
0
α
(
µs
(
∇u+ (∇u)T)
−2
3µs∇•u I
)
dt− α pI,
≈ ασ∗,olds,l + α∆t
(
µs
(
∇u+ (∇u)T)
−2
3µs∇•u I
)
− α pI. (34)
This equation contains the accumulating stress terms in u and the pressure
term, identical to the fluid stress formulation Equation 4. In the incompressible
fluid, ∇•u = 0, which fully aligns the terms
σf,l = µl
(
∇u+ (∇u)T)
− p I, (35)
12
The two equations are combined using Equation 17. The result can be inter-
preted as follows: while the fluid phase flows freely, the solid phase accumulates
displacement and stress, as indicated by the liquid fraction indicator.
σl = ασ∗,olds,l
+ (α∆t µs + (1− α)µl)(
∇u+ (∇u)T)
−2
3α∆t µs∇•u I
− p I.
(36)
Introducing combined viscosity νc:
νc = α∆t νs + (1− α) νl, (37)
the combined momentum equation reads:
∂u
∂t+∇•[(1− α)uu]
−∇•(νc[
∇u+ (∇u)T]
+2
3∇(νc ∇•u)
−∇•
[
νs(
∇d+ (∇d)T)]
+2
3∇(νs ∇•d)
= −1
ρ∇p,
(38)
where σolds,l has been expressed in terms of d. Note that the solid contribution
to νc contains the time-step size, which ultimately cancels out because of time-
integration of u into d.
The momentum equation splits the stress into the “old-time-level” contribu-
tion expressed in terms of accumulated displacement d and the implicit contri-165
bution depending on u. Explicit stress contribution is updated in accordance
with the d update, Equation 33.
The combined continuity equation reads:
∇•u = −α
K
∂p
∂t. (39)
To complete the solidification model, it is necessary to describe now the solid
fraction field α changes in the context of the energy equation and latent heat of
solidification, which will be described in the following section.170
13
5. Thermal Model
In this section, thermal aspects of the solidification model shall be explored.
This will include the thermal (buoyancy) effect in the fluid; a model for themally
induced stresses in the solid and a solidification model based on the release of
latent heat of solidification.175
5.1. Buoyancy Model in the Fluid
During solidification, there exists a significant temperature gradient within
the liquid phase, caused by intensive cooling on external surfaces and the heat
released during solidification at the fluid-solid interface. It is therefore desirable
to account for thermally-driven flow within the fluid flow model.180
The density change caused by the temperature gradient in liquids is typically
small compared to the mean density; further, the incompressibility assumption
stipulates that the liquid density remains constant, ie. it is not a function of
temperature. Under such conditions, it is sufficient to account for the thermal
driving force via the Boussinesq approximation [8]. The variation of density to
temperature is assumed to be linear around the reference density:
ρl(T ) = ρl,0 − βl(T − Tl,0), (40)
where T is the temperature, rhol,0 is the reference liquid density at reference
temperature Tl,0 and βl is the liquid thermal expansion coefficient.
The buoyancy body force appears in the liquid momentum equation Equation 8:
∂(ρl u)
∂t+∇•(ρl uu)−∇•(µl∇u) = −∇p+ [ρl,0 − βl(T − Tl,0)] g, (41)
where g is the gravitational acceleration.
This formulation is known to lead to numerical instability because the com-
putational molecule support for the two balancing terms (pressure gradient and
buoyancy) do not match. A transformation of the pressure into the dynamic
and quasi-hydrostatic component [9]:
pd = p− ρl(T )g•x, (42)
14
where x is the position vector, yielding a reformulated balanced momentum
source term:
−∇p+ ρl(T )g = −∇pd − (g•x)∇ρl(T ). (43)
Variation of ρl in other terms is neglected and the reference density in Equation 40
is chosen to match the mean liquid density:
ρl,0 = ρl. (44)
Dividing the equation set by ρl yields the final form of the fluid phase equations:
∂u
∂t+∇•(uu)−∇•(νl∇u) = −∇pd + g•x∇ [1− βl(T − Tl,0)] , (45)
∇•u = 0.
5.2. Thermal Stress Model in the Solid
Curiously, thermal stresses can be handled in a straightforward manner185
within the framework of the solid pressure and solid continuity equation. The
effect of heating on a solid is an increase in bulk volume (and thus a decrease
in solid density), expressed via the the thermal expansion coefficient αE .
In linear elasticity, thermally induced stress σs,l,T is added to the solid mo-
mentum equation:
σs,l,T = 3KαE(T − T0)I, (46)
where T0 is the reference temperature and K is precisely the bulk modulus from
Equation 21.190
This is clearly inconvenient for the stress formulation in terms of displace-
ment increment. As the spherical stress component is extracted into the solid
pressure, Equation 23, thermal stress can be accounted for as a change-of-
volume term in the solid continuity equation. First, in terms of solid displace-
ment d, Equation 23 is enhanced by the thermally induced stress:
∇•(ρsd) = −ρs
Kp+ 3KαE(T − T0), (47)
15
and then, taking the time derivative of Equation 47, in terms of solid velocity,
following Equation 26:
ρs
K
∂p
∂t+∇•(ρsu) = 3αE
∂T
∂t. (48)
Again, Equation 48 is linearised assuming that ∆ρs due to thermal expansion
is significantly smaller than ρs, as is the case with the Boussinesq hypothesis in
thermally driven liquid flows. Dividing Equation 48 by ρs yields:
∇•u = −1
K
∂p
∂t+ 3
αE
ρs
∂T
∂t. (49)
The thermal linear elastic solid is physically consistent and numerically con-
venient: any alternative formulation does not satisfy the integral volume in-
crease in the solid. Physically, the second term on r.h.s. of Equation 48 ac-
counts for the volumetric expansion of a heated solid and the resulting increase
in solid pressure, as a part of the solid continuity equation.195
With the above modification, the system is ready for the use in a blended
solid-fluid model, using the (p,u) as the primitive variables. To summarise, the
solid model consists of displacement accumulation (or solid velocity) equation
Equation 31, solid momentum, Equation 32 and thermally enhanced pressure
form of the solid continuity, Equation 49:
u =∂d
∂t,
∂u
∂t−∇•
[
νs∆t(
∇u+ (∇u)T)]
+2
3∇(νs∆t∇•u)
−∇•
[
νs(
∇d+ (∇d)T)]
+2
3∇(νs ∇•d)
= −1
ρ∇p,
(50)
∇•u = −1
K
∂p
∂t+ 3
αE
ρs
∂T
∂t.
This completes the formulation of a pressure-based linear elastic thermally
enhanced model for the solid. The model can be further developed towards the
thermal solid-fluid interaction; we shall however further enhance the model for
latent heat release and non-linearity in the solid stress model.
16
5.3. Phase Change Model200
During solidification, material initially existing in liquid state will solidify
under external cooling. The process of solidification will take place in three
phases:
• Evacuation of heat from a superheated liquid, described by the enegy
conservation in the fluid, Equation 3;205
• Process of phase change, during which the temperature of the melt re-
mains constant and the latent heat of phase change ∆H content decreases,
indicating phase change;
• Conductive cooling of the solidified material, described by the heat con-
duction equation in a solid, also described by Equation 3.210
During melting, the process is reversed.
The first and third phase are easily modelled within the existing framework.
The phase change process allows us to define the solid phase fraction increasing
from 0 to 1 as the evacuated heat changes from zero to ∆H [10].
With this definition, it is assumed that a phase change occurs in a pure215
single-component material. In metallurgical alloys, phase change occurs over
a temperature range and ∆H depends on the temperature, as opposed to the
step-function associated with isothermal solidification. Under such conditions,
a mush region appears, where the material is partially solid and partially liq-
uid. Numerical experiments show that a mushy region model with a narrow220
temperature range yields and stable and consistent formulation.
The simplest formulation of the solid fraction reads:
α =
1, T ≤ Tsol,
T−Tsol
Tliq−Tsol, Tsol < T < Tliq,
0, T ≥ Tliq,
(51)
where T is the temperature of the melt calculated from the energy equation and
Tliq and Tsol are the liquidus and solidus temperatures, chosen to straddle the
17
solidification temperature for the material TS by a small range ǫT :
Tsol = TS − ǫT , (52)
Tliq = TS + ǫT . (53)
However, this formulation yields to instability, as the latent heat release
drives changes in the temperature in an unstable manner. Further, while
Equation 51 is attractive, its physics is incomplete: the field of released la-
tent heat dH is also subject to convective transport if a mushy region exists225
and needs to be updated accordingly.
The problem is complicated by the fact that the combined specific heat and
thermal conductivity of the mushy region needs to be evaluated taking into
account the presence of both phases:
CP,c = αCP,s + (1− α)CP,l, (54)
where CP,c is the combined specific heat capacity of the mixture and CP,s and
CP,l are the solid and liquid specific heat capacities, respectively.
Equivalently, combined thermal conductivity kc is:
kc = αks + (1− α) kl, (55)
where ks and kl are the solid and liquid thermal conductivities.
Solidification model used in this study defines the solid fraction field α as a230
fraction of released latent heat dH of phase change compared to its total amount
∆H.
The initial distribution of dH is defined, based on the state of the melt at
the start of simulation:
dH = α∆H. (56)
At each time-step, dH is updated and limited for the range 0 ≤ dH ≤ ∆H. In
practice, the latent heat release must be limited, as follows:
dH = min
[
0,max
[
∆H, dH0 +
∫ τ=t
τ=0
cP,eff (T∗
− T )
]]
, (57)
= min[
0,max[
∆H, dHold + cP,eff (T∗
− T )]]
, (58)
18
where dH0 = dH(t = 0) and T ∗ is the reference temperature for latent heat
release, calculated as:
T ∗ = αTsol + (1− α)Tliq. (59)
Note that at the onset of solidification the reference temperature for heat release
is equal to Tliq, which gradually reduces to Tsol as solidification progresses.
Solid fraction can now be calculated from Equation 56:
α =dH
∆H. (60)
As dH is already bounded, bounds of α are automatically satisfied.235
The enthalpy source term Q is calculated from the dH distribution, ac-
counting for convective transport effects. In the solid, the convective transfer is
neglected, yielding convection fluxes which are not divergence-free. For reasons
of boundedness, non-conservative form of advection is chosen:
Q =∂dH
∂t+ u•∇dH. (61)
Note that in the isothermal case (ǫT = 0), with the absence of the mushy
region the convective part of the source terms in Equation 61 equals zero [10];
however, in the mushy region the convective transport needs to be accounted
for due to the presence of liquid phase in the mixture.
5.4. Simplified Form of the Energy Equation240
Under the conditions of incompressible flow, Equation 3 can be simplified by
neglecting the compression/expansion work p∇•u, work of the external forces
ρf•u, viscous stress work ∇•(σ•u) and the kinetic energy of the flow, on the
argument that the latent heat of phase change is orders of magnitude larger. In
the solid region, the convection term ∇•(ρeu) is also neglected, as u becomes245
small.
Heat conduction is modelled by Fourier’s law:
q = −κc∇T, (62)
19
where κc is the combined thermal conductivity:
κc = ακs + (1− α)κl (63)
and κs and κl are the thermal conductivities of the liquid and solid phase,
respectively.
Again, the convective transport in the solid shall be neglected for stability.
The final form of the energy equation is reached by dividing the energy
equation by ρCP in order to use temperature as the working variable. This
is beneficial as the temperature is a continuous variable across the liquid-solid
interface. Again, a non-conservative form of advection is chosen for boundedness
in the mushy region.
∂T
∂t+ (1− α)u•∇T −∇•
(
κc
ρc CP,c
∇T
)
=Q
ρc CP,c
. (64)
where rhoc, CP,c and κc are the combined density, specific heat capacity and250
thermal conductivity of the mixture, respectively. Further terms may arise if Cp
or κ are considered to vary with T or α. In this study, the effect of temperature
variation on thermal properties is neglected, as it is considered to be small
compared to other thermal effects.
6. Non-Linear Solid Phase Model Formulation255
A trivial solid mechanics model is a linear elastic model with constant mate-
rial properties and displacement d as the working variable [11] which has been
developed above into the form adequate for unified modelling in fluid-solid in-
teraction. This model is, however, not applicable to solidification, due to the
presence of material non-linearity (temperature- and solidification-dependent260
material properties) and geometric non-linearity (large deformations and rota-
tions) within the solidifying material. To remedy this, the solid stress model
formulation needs to be further enhanced.
It is interesting to notice that most non-linear stress models with sufficient
complexity to model the solidification process already operate in the so-called265
20
incremental formulation. Such models operate in terms of displacement and
stress increment, by necessity of integrating the stress and strain state in the
solid.
The stress model formulation used in this study is the Large Deformation
Stress Model in the incremental form, with the displacement increment δd cho-270
sen as a working variable [12]. The model is formulated as follows.
6.1. Working Variables in the Non-Linear Stress Model
The choice of the solid phase model is driven by the need to account for all
non-linearities in the system and at the same time align the formulation with
the fluid flow model. In fluids, the choice of working variables is naturally the275
(p,u) pair and we shall aim to achieve the same, following the developments for
the linear elastic stress model in section 3.
6.2. Stress-Strain Tensor Pair
The model shall operate on the stress-strain pair of the second Piola-Kirchoff
stress tensor increment and the Green-Lagrangian strain tensor increment.280
Second Piola-Kirchoff stress tensor increment δΣ is defined as:
δΣ = 2µsδE+ λs tr(δE)I, (65)
where µs and λs are the Lame’s coefficients.
Green-Lagrangian strain tensor increment δE is defined as:
δE =1
2
[
∇δd+ (∇δd)T +∇δd•(∇d)T +∇d•(∇δd)T +∇δd•(∇δd)T]
. (66)
In preparation for the combination of the fluid and solid model similar to
Equation 17, introducing δd = u∆t and writing δΣ the stress model can be
rewritten in terms of u, yielding the final form of the Green-Lagrangian strain
tensor increment:
δΣ = µs ∆t[
∇u+ (∇u)T +∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T)]
+1
2λs ∆t tr
[
∇u+ (∇u)T +∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T)]
I.
(67)
21
Integration formulas for displacement, strain and stress are used to accumulate
the solid stress as a function of displacement (increment) as before, and read:
d = d0 +
∫ t
0
u dt ≈ dold + u∆t ≈ dold +1
2(uold + unew)∆t, (68)
E = E0 +
∫ t
0
δE = Eold + δE, (69)
Σ = Σ0 +
∫ t
0
δΣ = Σold + δΣ, (70)
where d0, E0 and Σ0 represent the initial displacement, strain and stress. At
each time instance, the Cauchy stress σ in the solid can be recovered as:
σ =1
det(F)F•Σ•FT , (71)
where F is the deformation gradient tensor
F = I+ (∇d)T . (72)
6.3. Non-Linear Solid Pressure Term
Following the developments from subsection 3.3, the stress-strain pair from
subsection 6.2 will be reformulated to introduce the pressure term.
Comparing Equation 18, Equation 69 and Equation 70, it is clear that the285
non-linear stress can also be decomposed into the explicit “old-time-level” con-
tribution (σolds,l ) and the implicit terms in∇u. Identical terms exist in Equation 67
and the same procedure can be followed. This also applies to the pressure de-
composition of the stress term from subsection 3.4.
The modified form of the non-linear stress tensor increment reads:
δΣ = µs ∆t[
∇u+ (∇u)T +∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T)]
+1
2λs ∆t tr
[
∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T)]
I
−∆t
(
p+2
3µs ∇•u
)
.
(73)
The term µs ∆t[
∇u+ (∇u)T]
can be made implicit in u within the stress290
divergence in the momentum equation, as was the case in Equation 26, while
the newly introduced pressure term is subject to pressure-velocity coupling.
22
A further consequence of the decomposition is the linearised part of the solid
continuity equation, identical to the case of a linear elastic solid, Equation 26.
∇•(ρsu) = −ρs
K
∂p
∂t.
The definition of the bulk modulus K remains identical as before, Equation 21.
6.4. Non-Linear Solid Phase Equations
Combining the developments above and enhanced for the thermal stresses295
in the solid, the non-linear thermally enhanced stress model is summarised as
follows.
First, definition of the solid velocity Equation 31:
u =∂d
∂t
is used to accumulate displacement d from the velocity u, which is the primary
working variable, Equation 16.
dnew = dold +1
2(uold + unew)∆t.
Second, the non-linear solid momentum equation reads:
∂u
∂t−∇•
[
νs ∆t(
∇u+ (∇u)T)]
+2
3∇(νs∆t∇•u)
−∇•
[
νs ∆t(
∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T))]
−1
2∇•
[
λs ∆t tr(
∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T))
I]
−∇•Σold
= −1
ρ∇p.
(74)
This complicated expression requires clarification, by rows:
• The inertial term ∂u∂t; The convection term in the solid momentum is
equation neglected;300
• Linear part of stress increment, in terms of deformation velocity. This
term can be made implicit in u and matches the viscous shear stress in
the fluid, Equation 8;
23
• Non-linear part of shear stress increment, in combinations of d and u
terms, treated explicitly. For the linear solid stress model, this term does305
not exist;
• Non-linear part of the volumetric change, as above;
• Accumulated non-linear stress, treated explicitly. In the linear stress
model, Equation 32, this matches the explicit term
∇•
[
νs(
∇d+ (∇d)T)]
−2
3∇(νs ∇•d) (75)
and
• Pressure term, matching the pressure term in the liquid.
This is completed with pressure-based reconstruction of the non-linear stress
increment, Equation 73:
δΣ = µs ∆t[
∇u+ (∇u)T +∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T)]
+1
2λs ∆t tr
[
∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T)]
I
−∆t
(
p+2
3µs ∇•u
)
.
and, based on δΣ, the non-linear stress accumulation Equation 70:
Σnew = Σold + δΣ.
Finally, the thermally enhanced volumetric continuity Equation 49 reads:
∇•u = −1
K
∂p
∂t+ 3
αE
ρs
∂T
∂t.
Unsurprisingly, the non-linear solid stress model is a modest expansion on310
the linear form.
7. Non-Linear Solidification Model
Based on the above, the final form of the non-linear solid-fluid interaction,
and the non-liear solidification model follows the development from section 4
and subsection 5.3, using the solid formulation from subsection 6.4.315
24
Trapezoidal time integration for the solid displacement, Equation 33 is used
to accumulate the displacement d from the working variable u:
dnew = dold +1
2
(
αold uold + αnew unew)
∆t.
Combined non-linear mometum equation contains terms both in u and d,
where all terms in d consistently vanish in the liquid phase (α = 0):
∂u
∂t+∇•[(1− α)uu]−∇•
[
νc(
∇u+ (∇u)T)]
+2
3∇(νc ∇•u)
−∇•
[
ανs ∆t(
∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T))]
−1
2∇•
[
αλs ∆t tr(
∇u•(∇d)T +∇d•(∇u)T +∆t(
∇u•(∇u)T))
I]
−∇•
(
αΣold)
= −1
ρ∇p.
(76)
with the combined viscosity νc defined by Equation 37:
νc = α∆t νs + (1− α) νl.
Combined thermally enhanced continuity Equation 39:
∇•u = −α
K
∂p
∂t.
Combined (simplified) energy equation, Equation 64:
∂T
∂t+ (1− α)u•∇T −∇•
(
κc
ρc CP,c
∇T
)
=Q
ρc CP,c
.
with the phase change indicator α calculated from the latent heat fraction,
Equation 60.
α =dH
∆H
and the details of the phase change model are given in subsection 5.3. This
completes the formulation of the non-linear solidification model.
8. Solution Methodology
Looking at the unified equations of the solidification system, it is clear that
the nature of inter-equation coupling closely matches the fluid flow formulation,320
25
with additional non-linear explicit source terms in the momentum equation.
The formulation of the solution algorithms will be chosen accordingly.
The simplest algorithms dealing with pressure-velocity coupling are the SIM-
PLE [13] and PISO algorithms [14] and will be used in this study. Further op-
tions exist, eg. the block-coupled solution of the pressure-velocity system, which325
may be explored in the future.
While SIMPLE and PISO-type algorithms are we established in fluid dy-
namics, this is a relatively unusual choice of equation formulation and solution
for solid dynamics problems. It is therefore necessary to approach the validation
and verification of the new model in stages.330
IMPLEMENTATION NOTES: - solve velocity - create fluxes - solve pressure
- correct fluxes - use fluxes for div D terms - accumulate div D fluxes from div
U fluxes - BOUNDARY CONDITION CONSISTENCY
9. Basic Validation Cases
In the first instance, the combined model shall be tested for “pure” fluid335
flow and solid mechanics. As the formulation and solution algorithm originates
from the fluids and the equations natuarally reduce to incompressible fluid flow
equations for α = 0, flow validation cases are omitted. Emphasis shall be given
to static loading cases: in solidification, the only significant transient effect is
related to heat transfer.340
Wherever possible, simulation results shall be compared against analytical
or reference numerical solution.
9.1. Linear Elasticity
The first step in validation is a classic problem of stress concentration around
a circular hope loaded by uniform unidirectional tension, [15, 16, 11] HJ¡ HERE!!!345
26
9.2. Thermoelasticity
9.3. Simple Linear Fluid-Solid Interaction
9.4. Thermal Balance and Phase Change
9.5. Linear Solidification Modelling
9.6. Non-Linear Solidification Modelling350
10. Simulation of Solidification of an Alumunium Cast
For solid mechanics, basic tests have been performed for: linear and non-
linear elasticity and linear thermo-elasticity for cases with analytical solutions.
In parametric studies of mesh refinement, time-step size and discretisation set-
tings, the model performs perfectly across all cases, with examples of thermal355
expansion of a heated solid shown in Figure 1.
Conventional fluid-solid interaction cases can also be modelled without diffi-
culty. In such cases, there exists a step change in the fluid fraction indicator α,
delimiting the boundary between the fluid and solid. Examples of a travelling
pressure wave in an elastic pipe and a fluid jet hitting an elastic membrane are360
shown in Figure 2.
The model can now be considered ready for solidification applications, with
extensive validation and verification still in progress.
11. Conclusion and Future Work
This paper describes a combined model for fluid-solid interaction cases in the365
unified modelling approach, with the pressure p and velocity u chosen as the
working set of variables. The solid model accounts for material and geometric
non-linearities, include the novel handling of thermal stresses and is formulated
in terms of the momentum and continuity equation to avoid the Lame coefficient
singularity at Poisson’s ratio of ν = 0.5. The fluid model is a conventional370
single-phase transient laminar non-Newtonian flow model.
The blended fluid-solid model uses a fluid fraction variable α to indicate the
phase state and is capable of modelling the mushy zone solidification. At this
27
Figure 1: Combined FSI model for cases of thermal expansion of a solid brick and bar.
Figure 2: Conventional FSI cases: wave propagation in an elastic pipe and a jet impacting an
elastic membrane.
stage the combined model is validated in the cases of pure fluid flow, various
cases of linear and non-linear thermo-elasticity and simple fluid-solid interaction.375
The work towards a validated and verified solidification model continues and will
be reported in further publications.
28
References
[1] G. Hou, J. Wang, A. Layton, Numerical methods for fluid-structure inter-
action a review, Communications in Computational Physics 12 (2) (2012)380
337377. doi:https://doi.org/10.4208/cicp.291210.290411s.
[2] v. Tukovic, A. Karac, P. Cardiff, H. Jasak, A. Ivankovicc, Openfoam finite
volume solver for fluid-solid interaction, Transactions of FAMENA 42 (3)
(2018) 1–31. doi:https://doi.org/10.21278/TOF.42301.
[3] J. Degroote, J. Vierendeels, Multi-solver algorithms for the parti-385
tioned simulation of fluid-structure interaction, Computational Meth-
ods in Applied Mechanical Engineering 200 (CHECK) (2011) 21952210.
doi:https://doi.org/10.1016/j.cma.2011.03.015.
[4] C. Greenshields, H. Weller, A. Ivankovic, The finite volume method for
coupled fluid flow and stress analysis, Computer Modelling and Simulation390
in Engineering 4 (1999) 213–218.
[5] G. Papadakis, A novel pressure-velocity formulation and solution method
for fluid-structure interaction problems, Journal of Computational Physics
227 (6) (2008) 3383–3404. doi:CHECK.
[6] J. Degroote, K.-J. Bathe, J. Vierendeels, Performance of a new395
partitioned procedure versus a monolithic procedure in fluid-
structure interaction, Computers and Structures 87 (793801).
doi:https://doi.org/10.1016/j.compstruc.2008.11.013.
[7] T. Richter, A monolitic geometric multigrid solver for fluid-
structure interactions in ale formulation, International Jour-400
nal for Numerical Methods in Engineering 104 (2015) 372390.
doi:https://doi.org/10.1002/nme.4943.
[8] J. Boussinesq, Theorie de l’ecoulement tourbillant et tumultueux des liq-
uides dans les lits rectilignes a grande section, Vol. 16, Paris, Gauthier-
Villars et fils, 1897.405
29
[9] H. Jasak, Cfd analysis in subsea and marine technology, IOP Conference
Series: Materials Science and Engineering 276 (1) (2017) 012009.
URL http://stacks.iop.org/1757-899X/276/i=1/a=012009
[10] V. Voller, C. Prakash, A fixed grid numerical modelling methodology
for convection-diffusion mushy region phase-change problems, Interna-410
tional Journal of Heat and Mass Transfer 30 (8) (1987) 1709–1719.
doi:10.1016/0017-9310(87)90317-6.
[11] H. Jasak, H. Weller, Application of the finite volume method and unstruc-
tured meshes to linear elasticity, Int. J. Num. Meth. Engineering 48 (2)
(2000) 267–287.415
[12] P. Cardiff, Z. Tukovic, P. D. Jaeger, M. Clancy, A. Ivankovic, A Lagrangian
cell-centred finite volume method for metal forming simulation, Interna-
tional Journal for Numerical Methods in Engineering 109 (13) (2016) 1777–
1803. doi:10.1002/nme.5345.
[13] S. Patankar, D. Spalding, A calculation procedure for heat, mass and mo-420
mentum transfer in three-dimensional parabolic flows, Int. J. Heat Mass
Transfer 15 (1972) 1787.
[14] R. Issa, Solution of the implicitly discretized fluid flow equations by
operator-splitting, J. Comp. Physics 62 (1986) 40–65.
[15] I. Demirdzic, S. Muzaferija, Finite volume method for stress analysis in425
complex domains, Int. J. Numer. Meth. Engineering 37 (1994) 3751–3766.
[16] I. Demirdzic, S. Muzaferija, M. Peric, Benchmark solutions of some struc-
tural analysis problems using finite-volume method and multigrid acceler-
ation, Int. J. Numer. Meth. Engineering 40 (10) (1997) 1893–1908.
30
Recommended